Inspired by Large Language Models (LLMs), codec-based TTS systems have achieved remarkable success in high-fidelity speech generation by directly adopting the standard Cross-Entropy (CE) objective to predict sequences of discrete speech tokens from neural audio codecs. However, this practice overlooks the geometric structure of codec token spaces: tokens that are close in the codec codebook embedding space often correspond to acoustically similar speech realizations and can serve as perceptually plausible alternatives. We term this observation the Neighboring Token Phenomenon. The one-hot CE objective is agnostic to such structure, as it penalizes all non-target tokens equally, including these close neighbors, creating a mismatch between the structured codec latent space and the learning objective. In this paper, we quantitatively analyze the neighboring token phenomenon in both reconstruction and generation settings. To address the mismatch, we propose Multi-Neighbor Cross-Entropy (MNCE) loss function, which generalizes the one-hot target of CE loss into a positive set containing the ground-truth token and its nearest neighbors, encouraging the model to treat neighbor predictions as partially correct. As a result, even when the model misses the exact target token, it is more likely to predict a close neighbor rather than an unrelated token, reducing perceptually harmful errors in synthesized speech. Experiments demonstrate that MNCE is effective and generalizable across diverse codec-based TTS systems, improving quality and robustness. Moreover, scaling training to 20,000 hours further validates the scalability of our approach.

Reinforcement learning (RL) has become one of the primary paradigms for reasoning enhancement of large language models (LLMs). In particular, Group Relative Policy Optimization (GRPO) and related algorithms have demonstrated strong performance with outcome-level rewards. However, these methods depend solely on the final answer, without feedback regarding which intermediate steps contribute to success or failure. As task complexity and reasoning trajectory length increase, such sparse final-answer rewards become increasingly insufficient. To address this limitation, we introduce ConsensusBench, a novel dataset designed to provide rule-based process-level signals. We posit that a correct final answer relies on a small set of intermediate conclusions throughout the reasoning process, which can be seen as a verifiable sub-outcome. We identify these sub-outcomes by filtering correct trajectories from N rollouts and clustering semantically equivalent intermediate statements. We call these clustered statements as Consensus Nodes. By integrating a rule-based process reward derived from these nodes into GRPO-style algorithms, we develop a new reinforcement learning signal named ConsensusPR. It directly reduces the reward sparsity of outcome reward across long reasoning trajectories. To facilitate systematic process-level evaluation, we introduce three metrics to our benchmark: Final Answer Accuracy (Acc), Node Coverage Rate (NCR), and Tokens per Node (TPN). Experiments across AIME 2024, AIME 2025, GSM8K, MATH-500, and our ConsensusBench demonstrate that the proposed method consistently surpasses GRPO-style approaches, highlighting the practical value of consensus nodes in guiding reasoning.

Video-to-audio (V2A) generation aims to produce a single soundtrack that is simultaneously semantically faithful, temporally synchronized, aesthetically natural, and—for stereo output—spatially accurate. These perceptual axes are judged on the same waveform yet often pull generation in locally conflicting directions, while the dominant post-training recipe—reinforcement learning (RL) from an aggregated scalar reward—supervises them only at the final waveform with one entangled signal that gives little indication of which axis failed or where, so competing axes interfere through a single advantage value. We present MosaicAudio, a V2A framework that replaces terminal scalar rewards with dense, per-axis supervision distilled on-policy. MosaicAudio first trains four specialist flow-matching teachers, one per perceptual axis, then consolidates them into a single student: the student samples its own denoising trajectories and, at each visited state, receives a full teacher velocity field routed to one axis at a time, so that incompatible corrections are not averaged together. A perceptual manifold anchor keeps the student on a high-fidelity audio prior and curbs reward hacking during alignment. On the in-domain VGGSound test set and the out-of-domain AudioCanvas benchmark, MosaicAudio achieves state-of-the-art performance among compared V2A systems across objective metrics and human ratings, with especially consistent gains in complex multi-event scenes.

We propose UniVocal, a unified framework that implicitly infers vocal modes from text context to pioneer Speech-Singing Code-Switching (SCS) Synthesis—a task where transitions are autonomously driven by textual semantics, akin to seamless human language blending. Unlike single-mode generation or systems relying on switching-control tags, our proposed UniVocal implicitly infers vocal modes solely from text context. To achieve this, we employ a data-efficient two-stage curriculum learning strategy that progressively trains a competitive TTS system to acquire the desired SCS capability. Addressing data scarcity, we introduce a scalable pipeline to synthesize diverse code-switching data that is both semantically and acoustically natural, alongside a new multi-scenario benchmark, SCSBench. To address limitations of semantic tokenizers in capturing acoustic details, we also introduce refined cent token and Chain-of-Thought (CoT) generation for planning prosody before content generation, effectively enhancing empathetic speech generation and singing melody. Experimental results demonstrate that UniVocal achieves state-of-the-art performance on SCSBench while maintaining competitive performance on regular speech and singing tasks. Audio samples are available at https://project-univocal-demo.github.io/demo/. The code and dataset are released at https://github.com/FunAudioLLM/FunResearch/tree/main/UniVocal.

Large Language Models (LLMs) extend their capabilities through function-calling (FC), which relies on training data with high quality, diversity, and broad coverage of scenario. However, obtaining and annotating real function-calling data is challenging, while synthetic data from existing pipelines often suffers from unreliable APIs, limited tool scalability, insufficient diversity, and weak quality control. To address these, we present GenesisFunc, an automated pipeline for generating FC training data. Starting from reliable tools in widely used public benchmarks, our GenesisFunc employs a multi-agent framework to support a dialogue generation system that produces conversations spanning diverse scenarios, while maintaining both diversity and quality throughout the process. The accuracy of the data is further reinforced through a multi-stage evaluation system. We fine-tune an 8B LLM on the synthetic dataset and show through extensive experiments that it outperforms similarly sized open-source models in in-domain FC performance and out-of-domain generalization, while reaching FC capabilities comparable to some of the latest API-based models. In addition, our method demonstrates strong potential to scale effectively across downstream tools, underscoring its real-world applicability.

Although large language models (LLMs) excel in knowledge recall and reasoning, their static nature leads to outdated information as the real world evolves or when adapting to domain-specific knowledge, highlighting the need for effective knowledge injection. However, current research on knowledge injection remains superficial, mainly focusing on knowledge memorization and retrieval. This paper proposes a four-tier knowledge injection framework that systematically defines the levels of knowledge injection: memorization, retrieval, reasoning, and association. Based on this framework, we introduce DeepKnowledge, a synthetic experimental testbed designed for fine-grained evaluation of the depth of knowledge injection across three knowledge types (novel, incremental, and updated). We then explore various knowledge injection scenarios and evaluate the depth of knowledge injection for each scenario on the benchmark. Experimental results reveal key factors to reach each level of knowledge injection for LLMs and establish a mapping between the levels of knowledge injection and the corresponding suitable injection methods, aiming to provide a comprehensive approach for efficient knowledge injection across various levels. The code is available at https://github.com/icip-cas/Knowledge-Learning-Toolkits.

PCL detection task is aimed at identifying and categorizing language that is patronizing or condescending towards vulnerable communities in the general media. Compared to other NLP tasks of paragraph classification, the negative language presented in the PCL detection task is usually more implicit and subtle to be recognized, making the performance of common text classification approaches disappointed. Targeting the PCL detection problem in SemEval-2022 Task 4, in this paper, we give an introduction to our team’s solution, which exploits the power of prompt-based learning on paragraph classification. We reformulate the task as an appropriate cloze prompt and use pre2trained Masked Language Models to fill the cloze slot. For the two subtasks, binary classification and multi-label classification, DeBERTa model is adopted and fine-tuned to predict masked label words of task-specific prompts. On the evaluation dataset, for binary classification, our approach achieves an F1-score of 0.6406; for multi-label classification, our approach achieves an macro-F1-score of 0.4689 and ranks first in the leaderboard.

Machine Reading Comprehension with Unanswerable Questions is a difficult NLP task, challenged by the questions which can not be answered from passages. It is observed that subtle literal changes often make an answerable question unanswerable, however, most MRC models fail to recognize such changes. To address this problem, in this paper, we propose a span-based method of Contrastive Learning (spanCL) which explicitly contrast answerable questions with their answerable and unanswerable counterparts at the answer span level. With spanCL, MRC models are forced to perceive crucial semantic changes from slight literal differences. Experiments on SQuAD 2.0 dataset show that spanCL can improve baselines significantly, yielding 0.86 2.14 absolute EM improvements. Additional experiments also show that spanCL is an effective way to utilize generated questions.

This paper describes the DiDi AI Labs’ submission to the WMT2020 news translation shared task. We participate in the translation direction of Chinese->English. In this direction, we use the Transformer as our baseline model and integrate several techniques for model enhancement, including data filtering, data selection, back-translation, fine-tuning, model ensembling, and re-ranking. As a result, our submission achieves a BLEU score of 36.6 in Chinese->English.

Traditional topic models are insufficient for topic extraction in social media. The existing methods only consider text information or simultaneously model the posts and the static characteristics of social media. They ignore that one discusses diverse topics when dynamically interacting with different people. Moreover, people who talk about the same topic have different effects on the topic. In this paper, we propose an Interaction-Aware Topic Model (IATM) for microblog conversations by integrating network embedding and user attention. A conversation network linking users based on reposting and replying relationship is constructed to mine the dynamic user behaviours. We model dynamic interactions and user attention so as to learn interaction-aware edge embeddings with social context. Then they are incorporated into neural variational inference for generating the more consistent topics. The experiments on three real-world datasets show that our proposed model is effective.

Implicit discourse relation recognition aims to understand and annotate the latent relations between two discourse arguments, such as temporal, comparison, etc. Most previous methods encode two discourse arguments separately, the ones considering pair specific clues ignore the bidirectional interactions between two arguments and the sparsity of pair patterns. In this paper, we propose a novel neural Tensor network framework with Interactive Attention and Sparse Learning (TIASL) for implicit discourse relation recognition. (1) We mine the most correlated word pairs from two discourse arguments to model pair specific clues, and integrate them as interactive attention into argument representations produced by the bidirectional long short-term memory network. Meanwhile, (2) the neural tensor network with sparse constraint is proposed to explore the deeper and the more important pair patterns so as to fully recognize discourse relations. The experimental results on PDTB show that our proposed TIASL framework is effective.